Information processing method, device, electronic device and storage medium

Through the two-stage object detection model to identify and blur license plates and faces in panoramic images, the problems of low recognition accuracy and difficulty in privacy protection in the prior art are solved, and the effects of high accuracy and privacy protection are achieved.

CN114332924BActive Publication Date: 2024-12-06HEBEI DINGLIAN TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111555448.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-17
Publication Date
2024-12-06
Estimated Expiration
2041-12-17

AI Technical Summary

Technical Problem

When the prior art directly recognizes license plates or faces in panoramic images, it leads to a decrease in recognition accuracy and makes it difficult to effectively protect personal privacy.

Method used

Using a two-stage object detection model, firstly identify the target subject area in the image through the first target detection model, then identify the target object area in the target subject area through the second target detection model, and blur it to improve the recognition accuracy and privacy protection.

Benefits of technology

Accurately identifying target objects in smaller areas improves the accuracy of license plates and face recognition, while effectively protecting personal privacy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114332924B_ABST
    Figure CN114332924B_ABST
Patent Text Reader

Abstract

The present invention provides an information processing method, device, electronic device and storage medium, wherein the method comprises: obtaining an image to be processed; inputting the image to be processed into a first target detection model to obtain at least one target subject area output by the first target detection model; inputting at least one target subject area into a second target detection model to obtain a target object area in each target subject area; performing fuzzy processing on the target object in the image to be processed based on each target object area; wherein the first target detection model is obtained by training based on a plurality of first image samples containing the target subject, and the second target detection model is obtained by training based on a plurality of second image samples containing the target object. The information processing method, device, electronic device and storage medium provided by the present invention can identify the target object area in a smaller area, thereby improving the accuracy of target object recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an information processing method, device, electronic equipment and storage medium. Background Art

[0002] Panoramic images are spliced ​​after field framing, and have a realistic restoration performance. Combined with current computer technology, panoramic images with high definition and resolution can achieve all-round, interactive and real scenes. Therefore, panoramic images have a wide range of applications. With the development of network technology, panoramic images are easy to spread. The face and license plate elements in panoramic images are related to personal privacy. Therefore, the privacy issues involving face and license plate elements are becoming increasingly prominent.

[0003] In the related art, license plates or faces are usually directly recognized in panoramic images, and the detected license plates or faces are blurred.

[0004] However, in the above-mentioned related technologies, directly recognizing license plates or face pairs in panoramic images leads to reduced recognition accuracy. Summary of the invention

[0005] In view of the problems existing in the prior art, the present invention provides an information processing method, device, electronic device and storage medium.

[0006] The present invention provides an information processing method, comprising:

[0007] Acquire an image to be processed; wherein the image to be processed includes a target object to be processed;

[0008] Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model;

[0009] Inputting at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions;

[0010] Performing blur processing on the target object in the image to be processed based on each target object region;

[0011] The first target detection model is trained based on a plurality of first image samples containing target subjects, and the second target detection model is trained based on a plurality of second image samples containing target objects.

[0012] According to an information processing method provided by the present invention, the target object includes a target pedestrian; the first target detection model includes a pedestrian detection model, and the second target detection model includes a head detection model;

[0013] Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model includes:

[0014] Inputting the image to be processed into the pedestrian detection model to obtain at least one target pedestrian area output by the pedestrian detection model;

[0015] Inputting at least one of the target main regions into the second target detection model to obtain a target object region in each of the target main regions comprises:

[0016] Inputting at least one of the target pedestrian regions into a head detection model to obtain a target head region for each target pedestrian region output by the head detection model;

[0017] The blurring of the target object in the image to be processed based on each target object region comprises:

[0018] The target head in the image to be processed is blurred based on each target head region.

[0019] According to an information processing method provided by the present invention, inputting at least one target pedestrian region into a head detection model to obtain a target head region of each target pedestrian region output by the head detection model comprises:

[0020] Inputting at least one of the target pedestrian regions into a posture detection model to obtain a posture detection result of each target pedestrian region output by the posture detection model;

[0021] When it is determined that the posture detection result is that no human posture is detected, filtering out the corresponding target pedestrian area;

[0022] Inputting the remaining target pedestrian regions into the head detection model to obtain a target head region of each remaining target pedestrian region output by the head detection model;

[0023] The posture detection model is trained based on a plurality of third image samples containing pedestrian postures.

[0024] According to an information processing method provided by the present invention, the target object includes a target vehicle; the first target detection model includes a vehicle detection model, and the second target detection model includes a license plate position extraction model;

[0025] Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model includes:

[0026] Inputting the image to be processed into the vehicle detection model to obtain at least one target vehicle area output by the vehicle detection model;

[0027] Inputting at least one of the target main regions into the second target detection model to obtain a target object region in each of the target main regions comprises:

[0028] Inputting at least one of the target vehicle regions into a license plate position extraction model to obtain a target license plate region for each target vehicle region output by the license plate position extraction model;

[0029] The blurring of the target object in the image to be processed based on each target object region comprises:

[0030] The target license plate in the image to be processed is blurred based on each target license plate area.

[0031] According to an information processing method provided by the present invention, the blurring of the target license plate in the image to be processed based on each target license plate area comprises:

[0032] Inputting at least one of the target license plate regions into a license plate recognition model to obtain a correction result for each target license plate region output by the license plate recognition model;

[0033] When it is determined that the correction result is that the license plate is not detected, filtering out the corresponding target license plate area;

[0034] Performing blur processing on the target license plate in the image to be processed based on each remaining target license plate area;

[0035] Among them, the license plate recognition model is obtained by training based on a fourth image sample; the fourth image sample includes license plate images of each license plate at multiple different angles, and the multiple license plate images at different angles include a front view license plate image.

[0036] According to an information processing method provided by the present invention, before inputting at least one of the target license plate regions into a license plate recognition model to obtain a correction result of each target license plate region output by the license plate recognition model, the method further includes:

[0037] Performing angle correction on at least one of the target license plate regions;

[0038] The step of inputting at least one of the target license plate regions into a license plate recognition model to obtain a correction result of each target license plate region output by the license plate recognition model includes:

[0039] At least one angle-corrected target license plate region is input into the license plate recognition model to obtain a correction result of each angle-corrected target license plate region output by the license plate recognition model.

[0040] The present invention also provides an information processing device, comprising:

[0041] An acquisition unit, used for acquiring an image to be processed; wherein the image to be processed includes a target object to be processed;

[0042] A first detection unit, configured to input the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model;

[0043] A second detection unit, configured to input at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions;

[0044] A processing unit, configured to perform blur processing on the target object in the image to be processed based on each target object region;

[0045] The first target detection model is trained based on a plurality of first image samples containing target subjects, and the second target detection model is trained based on a plurality of second image samples containing target objects.

[0046] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above-mentioned information processing methods when executing the program.

[0047] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned information processing methods are implemented.

[0048] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of any one of the above-mentioned information processing methods are implemented.

[0049] The present invention provides an information processing method, device, electronic device and storage medium, which firstly identifies the target subject area in the image to be processed by a first target detection model, then identifies the target object area in each target subject area by a second target detection model, and finally performs fuzzy processing on the target object in the image to be processed based on the identified target object area. In this way, the target object area can be identified in a smaller area, thereby improving the accuracy of target object identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 It is one of the flowcharts of the information processing method provided by the present invention;

[0052] Figure 2 This is the second flow chart of the information processing method provided by the present invention;

[0053] Figure 3 This is the third flow chart of the information processing method provided by the present invention;

[0054] Figure 4 This is the fourth flow chart of the information processing method provided by the present invention;

[0055] Figure 5 This is the fifth flow chart of the information processing method provided by the present invention;

[0056] Figure 6 This is the sixth flow chart of the information processing method provided by the present invention;

[0057] Figure 7 It is a structural schematic diagram of the information processing device provided by the present invention;

[0058] Figure 8 It is a schematic diagram of the physical structure of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] Combine the following Figure 1-Figure 6 The information processing method of the present invention is described.

[0061] Figure 1 It is one of the flowcharts of the information processing method provided by the present invention, such as Figure 1 As shown, the information processing method includes:

[0062] Step 101: Obtain an image to be processed.

[0063] The image to be processed includes a target object to be processed, and the image to be processed may be a panoramic image.

[0064] For example, the image to be processed may be obtained from a storage device; the target scene may be photographed in real time to obtain the image to be processed; or the image to be processed may be received from other devices.

[0065] Step 102: input the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model.

[0066] The first target detection model is trained based on a plurality of first image samples containing the target subject.

[0067] For example, when an image to be processed is obtained, the image to be processed is input into a pre-trained first target detection model, and the first target detection model recognizes the target subject features in the image to be processed to obtain at least one identified target subject area. The specific target subject area can be an area including pedestrians or an area including vehicles.

[0068] Step 103: input at least one of the target main regions into a second target detection model to obtain a target object region in each of the target main regions.

[0069] The second target detection model is trained based on a plurality of second image samples containing target objects.

[0070] For example, when at least one target body area is identified, the at least one target body area is input into the second target detection model, and the second target detection model further identifies the features of the target object in each target body area to obtain the target object area in each identified target area. The target object area can be a target license plate area or a target head area.

[0071] Step 104: perform blur processing on the target object in the image to be processed based on each target object region.

[0072] For example, when each target license plate area and / or each target head area is identified, the corresponding target license plate area is determined in the image to be processed based on the position information of each target license plate area, and the target license plate area is blurred. The specific blurring method may be marking each target license plate area with a mosaic, or erasing each target license plate area, or filling each target license plate area with a color that can cover the license plate information, etc.; the corresponding target head area is determined in the image to be processed based on the position information of each target head area, and the target head area is blurred. The specific blurring method may be marking each target head area with a mosaic, or erasing each target head area, or filling each target head area with a color that can cover the head information, etc.

[0073] The information processing method provided by the present invention firstly identifies the target subject area in the image to be processed by a first target detection model, then identifies the target object area in each target subject area by a second target detection model, and finally performs fuzzy processing on the target object in the image to be processed based on the identified target object area. In this way, the target object area can be identified in a smaller area, thereby improving the accuracy of target object identification.

[0074] Optionally, the target object includes a target pedestrian; the first target detection model includes a pedestrian detection model, and the second target detection model includes a head detection model. Figure 2 This is the second flow chart of the information processing method provided by the present invention, such as Figure 2 As shown, Figure 1 Step 102 in the embodiment specifically comprises the following steps:

[0075] 1021. Input the image to be processed into the pedestrian detection model to obtain at least one target pedestrian area output by the pedestrian detection model.

[0076] For example, when an image to be processed containing pedestrians is obtained, the image to be processed is input into a pre-trained pedestrian detection model, and the pedestrian detection model identifies the pedestrian features (for example, leg features, head features, arm features, etc.) in the image to be processed, and finally obtains at least one target pedestrian area. The information of the target pedestrian area may include the size and position of the target pedestrian area, etc.

[0077] In addition, the specific training method of the pedestrian detection model is as follows:

[0078] Collect multiple image samples containing pedestrians, then mark the pedestrian area in each image sample containing pedestrians, input all the marked image samples containing pedestrians into an initial pedestrian detection model (for example, a YOLO model) to obtain the pedestrian area predicted by the initial pedestrian detection model; then construct a first loss function based on the similarity between the predicted pedestrian area and the marked real pedestrian area, optimize the model parameters of the initial pedestrian detection model based on the first loss function until the convergence condition is reached, and obtain a trained pedestrian detection model.

[0079] like Figure 2 As shown, Figure 1 Step 103 in the embodiment specifically comprises the following steps:

[0080] Step 1031: input at least one of the target pedestrian regions into a head detection model to obtain a target head region for each target pedestrian region output by the head detection model.

[0081] For example, when at least one target pedestrian area is identified, the at least one target pedestrian area is input into a pre-trained head detection model, and the head detection model identifies the head features (for example, facial features, head contour features, and hair features, etc.) in each target pedestrian area, and finally obtains the target head area in each target pedestrian area. The information of the target head area may include the size and position of the target head area, etc.

[0082] In addition, the specific training method of the head detection model is as follows:

[0083] A plurality of image samples containing the head are collected, and then the head region in each image sample containing the head is marked, and all the marked image samples containing the head are input into an initial head detection model (for example, a YOLO model) to obtain the head region predicted by the initial head detection model; then a second loss function is constructed based on the similarity between the predicted head region and the marked real head region, and the model parameters of the initial head detection model are optimized based on the second loss function until the convergence condition is reached, thereby obtaining a trained head detection model.

[0084] like Figure 2 As shown, Figure 1 Step 104 in the embodiment specifically comprises the following steps:

[0085] Step 1041: perform blur processing on the target head in the image to be processed based on each target head region.

[0086] For example, when each target head region is acquired, the corresponding target head region is determined in the image to be processed according to the position information and size information of each target head region, and the target head region is blurred.

[0087] The information processing method provided by the present invention firstly identifies the target pedestrian region in the image to be processed by using a pedestrian detection model, then identifies the target head region in each target pedestrian region by using a head detection model, and finally performs fuzzy processing on the target head in the image to be processed based on the identified target head region. In this way, the target head region can be identified in a smaller area, thereby improving the accuracy of head recognition.

[0088] Optionally, Figure 3 This is the third flow chart of the information processing method provided by the present invention, such as Figure 3 As shown, Figure 2 Step 1031 in the embodiment specifically includes the following steps:

[0089] Step 10311: input at least one of the target pedestrian regions into a posture detection model to obtain a posture detection result for each target pedestrian region output by the posture detection model.

[0090] The posture detection model is trained based on a plurality of third image samples containing pedestrian postures.

[0091] For example, when at least one target pedestrian area is identified, the at least one target pedestrian area is input into a pre-trained posture detection model, and the posture detection model identifies the posture features (such as leg posture features, arm posture features, and foot posture features, etc.) in each target pedestrian area. If the posture detection model identifies the posture features in the target pedestrian area, the posture detection result of the target pedestrian area is output as detected human body posture; if the posture detection model does not identify the posture features in the target pedestrian area, the posture detection result of the target pedestrian area is output as no human body posture is detected.

[0092] In addition, the specific training method of the posture detection model is as follows:

[0093] Collect multiple image samples containing pedestrian postures, then mark the pedestrian posture in each image sample containing the pedestrian posture, input all the marked image samples containing the pedestrian posture into an initial posture detection model (for example, a YOLO model), and obtain the posture detection result predicted by the initial posture detection model; then construct a third loss function based on the similarity between the predicted posture detection result and the actual result of the marked pedestrian posture, and optimize the model parameters of the initial posture detection model based on the third loss function until the convergence condition is reached, thereby obtaining a trained posture detection model.

[0094] Step 10312: When it is determined that the posture detection result is that no human posture is detected, filter out the corresponding target pedestrian area.

[0095] For example, when the posture detection result of each target pedestrian area is obtained, the posture detection result of each target pedestrian area is analyzed. When it is determined that the posture detection result is that no human posture is detected, it means that the target pedestrian area is not a real pedestrian area. At this time, the information of the corresponding target pedestrian area is deleted; when it is determined that the posture detection result is that a human posture is detected, it means that the target pedestrian area is a real pedestrian area. At this time, the information of the corresponding target pedestrian area is retained, thereby achieving filtering of at least one target pedestrian area.

[0096] Step 10313: input the remaining target pedestrian regions into the head detection model to obtain the target head region of each remaining target pedestrian region output by the head detection model.

[0097] For example, after filtering out at least one target pedestrian area that is not a real pedestrian area, the remaining target pedestrian area is input into a head detection model, and the head detection model identifies the head features in the remaining target pedestrian area, and finally obtains the target head area of ​​the remaining target pedestrian area; then, based on the position information and size information of the target head area of ​​the remaining target pedestrian area, the corresponding target head area is determined in the image to be processed, and the target head area is blurred.

[0098] The information processing method provided by the present invention filters out non-real pedestrian areas in at least one target pedestrian area through a posture detection model, which can reduce the burden of data processing by the head detection model and also improve the accuracy of head privacy processing.

[0099] Optionally, the target object includes a target vehicle; the first target detection model includes a vehicle detection model, and the second target detection model includes a license plate position extraction model. Figure 4 This is a fourth flow chart of the information processing method provided by the present invention, such as Figure 4 As shown, Figure 1 Step 102 in the embodiment specifically comprises the following steps:

[0100] Step 1022: input the image to be processed into the vehicle detection model to obtain at least one target vehicle area output by the vehicle detection model.

[0101] For example, when an image to be processed containing a vehicle is obtained, the image to be processed is input into a pre-trained vehicle detection model, and the vehicle detection model identifies the vehicle features (for example, vehicle contour features, wheel features, and headlight features, etc.) in the image to be processed, and finally obtains at least one target vehicle area. The information of the target vehicle area may include the size and position of the target vehicle area, etc.

[0102] In addition, the specific training method of the vehicle detection model is as follows:

[0103] Collect multiple image samples containing vehicles, then mark the vehicle area in each image sample containing vehicles, input all the marked image samples containing vehicles into the initial vehicle detection model (MASK-RCNN model), and obtain the vehicle area predicted by the initial vehicle detection model; then construct a fourth loss function based on the similarity between the predicted vehicle area and the marked real vehicle area, and optimize the model parameters of the initial vehicle detection model based on the fourth loss function until the convergence condition is reached, thereby obtaining a trained vehicle detection model.

[0104] like Figure 4 As shown, Figure 1 Step 103 in the embodiment specifically comprises the following steps:

[0105] Step 1032: input at least one of the target vehicle regions into a license plate position extraction model to obtain a target license plate region for each target vehicle region output by the license plate position extraction model.

[0106] For example, when at least one target vehicle area is identified, the at least one target vehicle area is input into a pre-trained license plate position extraction model, and the license plate position extraction model identifies the license plate features (for example, license plate shape features and license plate alphanumeric combination features, etc.) in each target vehicle area, and finally obtains the target license plate area in each target vehicle area. The information of the target license plate area may include the size and position of the target license plate area, etc.

[0107] In addition, the specific training method of the license plate position extraction model is as follows:

[0108] Collect multiple image samples containing license plates, then mark the license plate area in each image sample containing the license plate, input all the marked image samples containing the license plate into the initial license plate position extraction model, and obtain the license plate area predicted by the initial license plate position extraction model; then construct a fifth loss function based on the similarity between the predicted license plate area and the marked real license plate area, and optimize the model parameters of the initial license plate position extraction model based on the fifth loss function until the convergence condition is reached, thereby obtaining a trained license plate position extraction model.

[0109] like Figure 4 As shown, Figure 1 Step 104 in the embodiment specifically comprises the following steps:

[0110] Step 1042: blur the target license plate in the image to be processed based on each target license plate region.

[0111] For example, when each target license plate area is acquired, the corresponding target license plate area is determined in the image to be processed according to the position information and size information of each target license plate area, and the target license plate area is blurred.

[0112] The information processing method provided by the present invention firstly identifies the target vehicle area in the image to be processed by the vehicle detection model, then identifies the target license plate area in each target vehicle area by the license plate position extraction model, and finally performs fuzzy processing on the target license plate in the image to be processed based on the identified target license plate area. In this way, the target license plate area can be identified in a smaller area, thereby improving the accuracy of license plate recognition.

[0113] Optionally, Figure 5 It is a flow chart of the information processing method provided by the present invention, such as Figure 5 As shown, Figure 4 Step 1042 in the embodiment specifically includes the following steps:

[0114] Step 10421: input at least one of the target license plate regions into a license plate recognition model to obtain a correction result for each target license plate region output by the license plate recognition model.

[0115] Among them, the license plate recognition model is obtained by training based on a fourth image sample; the fourth image sample includes license plate images of each license plate at multiple different angles, and the multiple license plate images at different angles include a front view license plate image.

[0116] For example, since the license plate position extraction model may identify multiple target license plate areas for a vehicle, when at least one target license plate area for each target vehicle area is identified, the at least one target license plate area is input into a pre-trained license plate recognition model, and the license plate recognition model identifies the license plate content features in each target license plate area. If the license plate recognition model recognizes the license plate content features in the target license plate area, the correction result of the target license plate area is output as detected license plate; if the license plate recognition model does not recognize the license plate content features in the target license plate area, the correction result of the target license plate area is output as undetected license plate.

[0117] In addition, the specific training method of the license plate recognition model is as follows:

[0118] For each license plate, multiple license plate images at different angles are collected as fourth image samples, and the multiple license plate images at different angles include a frontal license plate image, and then the license plate area of ​​the license plate image at each angle in the fourth image sample is marked, and the marked fourth image sample is input into an initial license plate recognition model (for example, a YOLO model) to obtain the license plate area predicted by the initial license plate recognition model; then a sixth loss function is constructed based on the similarity between the predicted license plate area and the marked real license plate area, and the model parameters of the initial license plate recognition model are optimized based on the sixth loss function until the convergence condition is reached to obtain a trained license plate recognition model. Since the license plate recognition model is trained for license plate images at different angles for each license plate, the recognition accuracy of the license plate recognition model is relatively high compared to the above-mentioned license plate position extraction model.

[0119] Step 10422: When it is determined that the correction result is that the license plate is not detected, filter out the corresponding target license plate area.

[0120] For example, when the correction result of each target license plate area is obtained, the correction result of each target license plate area is analyzed. When it is determined that the correction result is that the license plate is not detected, it means that the target license plate area is not a real license plate area. At this time, the information of the corresponding target license plate area is deleted; when it is determined that the correction result is that the license plate is detected, it means that the target license plate area is a real license plate area. At this time, the information of the corresponding target license plate area is retained, thereby achieving filtering of at least one target license plate area.

[0121] Step 10423: blur the target license plate in the image to be processed based on each remaining target license plate area.

[0122] For example, after filtering out at least one target license plate area that is not a real license plate area, the corresponding target license plate area is determined in the image to be processed based on the position information and size information of the remaining target license plate areas, and the determined target license plate area is blurred.

[0123] The information processing method provided by the present invention filters the non-real license plate areas in at least one target license plate area through a license plate recognition model, which can reduce the burden of the license plate recognition model in processing data and also improve the accuracy of license plate privacy processing.

[0124] Optionally, Figure 6 This is the sixth flow chart of the information processing method provided by the present invention, such as Figure 6 As shown, in the execution Figure 5 Before step 10421 in the method, the method further includes the following steps:

[0125] Step 10424: perform angle correction on at least one of the target license plate areas.

[0126] For example, when obtaining at least one target license plate area, since the license plate area may be the area corresponding to a tilted license plate, each target license plate area needs to be angle corrected. The specific angle correction method may be: inputting at least one target license plate area into a pre-trained image geometry correction model to obtain each angle-corrected target license plate area output by the image geometry correction model.

[0127] It should be noted that the angle correction of each target license plate area can also be performed based on the edge detection method, and there is no limitation to this.

[0128] Further, Figure 5 Step 10421 in the above method can be implemented in the following way:

[0129] At least one angle-corrected target license plate region is input into the license plate recognition model to obtain a correction result of each angle-corrected target license plate region output by the license plate recognition model.

[0130] In addition, the specific training method of the image geometry correction model is as follows:

[0131] A plurality of license plate images at different angles are collected as image samples, and then the angle of each license plate image in the image samples is marked, and the marked image samples are input into the initial image geometry correction model (for example, the YOLO model) to obtain the prediction result output by the initial image geometry correction model; then a seventh loss function is constructed based on the similarity between the prediction result and the marked true angle, and the model parameters of the initial image geometry correction model are optimized based on the seventh loss function until the convergence condition is reached, thereby obtaining a trained image geometry correction model.

[0132] The information processing method provided by the present invention performs angle correction on each target license plate area and then determines the correction result of each angle-corrected target license plate area to improve the accuracy of the correction result.

[0133] The information processing device provided by the present invention is described below. The information processing device described below and the information processing method described above can be referenced to each other.

[0134] Figure 7 is a schematic diagram of the structure of the information processing device provided by the present invention, such as Figure 7 As shown, the information processing device includes an acquisition unit 701, a first detection unit 702, a second detection unit 703 and a processing unit 704; wherein:

[0135] The acquisition unit 701 is used to acquire an image to be processed; wherein the image to be processed includes a target object to be processed;

[0136] A first detection unit 702 is used to input the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model;

[0137] A second detection unit 703 is used to input at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions;

[0138] A processing unit 704 is used to perform blur processing on the target object in the image to be processed based on each target object region;

[0139] The first target detection model is trained based on a plurality of first image samples containing target subjects, and the second target detection model is trained based on a plurality of second image samples containing target objects.

[0140] The information processing device provided by the present invention first identifies the target subject area in the image to be processed by a first target detection model, then identifies the target object area in each target subject area by a second target detection model, and finally performs blur processing on the target object in the image to be processed based on the identified target object area. In this way, the target object area can be identified in a smaller area, thereby improving the accuracy of target object identification.

[0141] Based on any of the foregoing embodiments, the target object includes a target pedestrian; the first target detection model includes a pedestrian detection model, and the second target detection model includes a head detection model; the first detection unit 702 is specifically used to:

[0142] Inputting the image to be processed into the pedestrian detection model to obtain at least one target pedestrian area output by the pedestrian detection model;

[0143] The second detection unit 703 is specifically used for:

[0144] Inputting at least one of the target pedestrian regions into a head detection model to obtain a target head region for each target pedestrian region output by the head detection model;

[0145] The processing unit 704 is specifically used for:

[0146] The target head in the image to be processed is blurred based on each target head region.

[0147] Based on any of the foregoing embodiments, the second detection unit 703 is further specifically configured to:

[0148] Inputting at least one of the target pedestrian regions into a posture detection model to obtain a posture detection result of each target pedestrian region output by the posture detection model;

[0149] When it is determined that the posture detection result is that no human posture is detected, filtering out the corresponding target pedestrian area;

[0150] Inputting the remaining target pedestrian regions into the head detection model to obtain a target head region of each remaining target pedestrian region output by the head detection model;

[0151] The posture detection model is trained based on a plurality of third image samples containing pedestrian postures.

[0152] Based on any of the above embodiments, the target object includes a target vehicle; the first target detection model includes a vehicle detection model, and the second target detection model includes a license plate position extraction model; the first detection unit 702 is further specifically used for:

[0153] Inputting the image to be processed into the vehicle detection model to obtain at least one target vehicle area output by the vehicle detection model;

[0154] The second detection unit 703 is further specifically configured to:

[0155] Inputting at least one of the target vehicle regions into a license plate position extraction model to obtain a target license plate region for each target vehicle region output by the license plate position extraction model;

[0156] The processing unit 704 is further specifically configured to:

[0157] The target license plate in the image to be processed is blurred based on each target license plate area.

[0158] Based on any of the foregoing embodiments, the processing unit 704 is further specifically configured to:

[0159] Inputting at least one of the target license plate regions into a license plate recognition model to obtain a correction result for each target license plate region output by the license plate recognition model;

[0160] When it is determined that the correction result is that the license plate is not detected, filtering out the corresponding target license plate area;

[0161] Performing blur processing on the target license plate in the image to be processed based on each remaining target license plate area;

[0162] Among them, the license plate recognition model is obtained by training based on a fourth image sample; the fourth image sample includes license plate images of each license plate at multiple different angles, and the multiple license plate images at different angles include a front view license plate image.

[0163] Based on any of the foregoing embodiments, the processing unit 704 is further specifically configured to:

[0164] Performing angle correction on at least one of the target license plate regions;

[0165] At least one angle-corrected target license plate region is input into the license plate recognition model to obtain a correction result of each angle-corrected target license plate region output by the license plate recognition model.

[0166] Figure 8 is a schematic diagram of the physical structure of the electronic device provided by the present invention, such as Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the information processing method, which includes: acquiring an image to be processed; wherein the image to be processed includes a target object to be processed;

[0167] Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model;

[0168] Inputting at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions;

[0169] Performing blur processing on the target object in the image to be processed based on each target object region;

[0170] The first target detection model is trained based on a plurality of first image samples containing target subjects, and the second target detection model is trained based on a plurality of second image samples containing target objects.

[0171] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0172] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored in a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the information processing method provided by the above methods, the method includes: obtaining an image to be processed; wherein the image to be processed includes a target object to be processed;

[0173] Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model;

[0174] Inputting at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions;

[0175] Performing blur processing on the target object in the image to be processed based on each target object region;

[0176] The first target detection model is trained based on a plurality of first image samples containing target subjects, and the second target detection model is trained based on a plurality of second image samples containing target objects.

[0177] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to perform the information processing method provided by the above methods, the method comprising: acquiring an image to be processed; wherein the image to be processed includes a target object to be processed;

[0178] Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model;

[0179] Inputting at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions;

[0180] Performing blur processing on the target object in the image to be processed based on each target object region;

[0181] The first target detection model is trained based on a plurality of first image samples containing target subjects, and the second target detection model is trained based on a plurality of second image samples containing target objects.

[0182] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0183] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An information processing method, characterized in that: include: Acquire an image to be processed; wherein the image to be processed includes a target object to be processed, and the image to be processed is a panoramic image; Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model; Inputting at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions; Performing blur processing on the target object in the image to be processed based on each target object region; The first target detection model is obtained by training based on a plurality of first image samples containing target subjects, and the second target detection model is obtained by training based on a plurality of second image samples containing target objects; In the case where the target object includes a target pedestrian, the first target detection model includes a pedestrian detection model, and the second target detection model includes a head detection model; Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model includes: Inputting the image to be processed into the pedestrian detection model to obtain at least one target pedestrian area output by the pedestrian detection model; Inputting at least one of the target main regions into the second target detection model to obtain a target object region in each of the target main regions comprises: Inputting at least one of the target pedestrian regions into a posture detection model to obtain a posture detection result of each target pedestrian region output by the posture detection model; When it is determined that the posture detection result is that no human posture is detected, filtering out the corresponding target pedestrian area; Inputting the remaining target pedestrian regions into the head detection model to obtain a target head region of each remaining target pedestrian region output by the head detection model; wherein the posture detection model is trained based on a plurality of third image samples containing pedestrian postures; The blurring of the target object in the image to be processed based on each target object region comprises: The target head in the image to be processed is blurred based on each target head region.

2. The information processing method according to claim 1, characterized in that: In the case where the target object includes a target vehicle, the first target detection model includes a vehicle detection model, and the second target detection model includes a license plate position extraction model; Inputting the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model includes: Inputting the image to be processed into the vehicle detection model to obtain at least one target vehicle area output by the vehicle detection model; Inputting at least one of the target main regions into the second target detection model to obtain a target object region in each of the target main regions comprises: Inputting at least one of the target vehicle regions into a license plate position extraction model to obtain a target license plate region for each target vehicle region output by the license plate position extraction model; The blurring of the target object in the image to be processed based on each target object region comprises: The target license plate in the image to be processed is blurred based on each target license plate area.

3. The information processing method according to claim 2, characterized in that: The blurring of the target license plate in the image to be processed based on each target license plate area includes: Inputting at least one of the target license plate regions into a license plate recognition model to obtain a correction result for each target license plate region output by the license plate recognition model; When it is determined that the correction result is that the license plate is not detected, filtering out the corresponding target license plate area; Performing blur processing on the target license plate in the image to be processed based on each remaining target license plate area; Among them, the license plate recognition model is obtained by training based on a fourth image sample; the fourth image sample includes license plate images of each license plate at multiple different angles, and the multiple license plate images at different angles include a front view license plate image.

4. The information processing method according to claim 3, characterized in that: Before inputting at least one of the target license plate regions into the license plate recognition model to obtain a correction result of each target license plate region output by the license plate recognition model, the method further includes: Performing angle correction on at least one of the target license plate regions; The step of inputting at least one of the target license plate regions into a license plate recognition model to obtain a correction result of each target license plate region output by the license plate recognition model includes: At least one angle-corrected target license plate region is input into the license plate recognition model to obtain a correction result of each angle-corrected target license plate region output by the license plate recognition model.

5. An information processing device, characterized in that: include: An acquisition unit, used for acquiring an image to be processed; wherein the image to be processed includes a target object to be processed, and the image to be processed is a panoramic image; A first detection unit, configured to input the image to be processed into a first target detection model to obtain at least one target main area output by the first target detection model; A second detection unit, configured to input at least one of the target body regions into a second target detection model to obtain a target object region in each of the target body regions; A processing unit, configured to perform blur processing on the target object in the image to be processed based on each target object region; The first target detection model is obtained by training based on a plurality of first image samples containing target subjects, and the second target detection model is obtained by training based on a plurality of second image samples containing target objects; In the case where the target object includes a target pedestrian, the first target detection model includes a pedestrian detection model, and the second target detection model includes a head detection model; The first detection unit is specifically used for: Inputting the image to be processed into the pedestrian detection model to obtain at least one target pedestrian area output by the pedestrian detection model; The second detection unit is specifically used for: Inputting at least one of the target pedestrian regions into a posture detection model to obtain a posture detection result of each target pedestrian region output by the posture detection model; When it is determined that the posture detection result is that no human posture is detected, filtering out the corresponding target pedestrian area; Inputting the remaining target pedestrian regions into the head detection model to obtain a target head region of each remaining target pedestrian region output by the head detection model; wherein the posture detection model is trained based on a plurality of third image samples containing pedestrian postures; The processing unit is specifically used for: The target head in the image to be processed is blurred based on each target head region.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the information processing method according to any one of claims 1 to 4 are implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the information processing method according to any one of claims 1 to 4 are implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the information processing method according to any one of claims 1 to 4 are implemented.